CodeLensAI

Read code like a map. Paste a function and CodeLensAI gives you three things at once:

1. A plain-English, line-by-line breakdown of what the code does.
2. An AI-written summary plus a time-complexity estimate (Big-O).
3. A flowchart of the control flow, rendered with Mermaid.

It supports Python, JavaScript, and TypeScript, runs entirely on free infrastructure, and needs no API key to work.

Live: https://codelensai-zeta.vercel.app

---

How it works

frontend/ (React + Vite)  ──fetch──>  api/explain.py  (Vercel Python function)
                                            │
                                            ├─ _lib/parser.py     Python -> IR (via ast)
                                            ├─ _lib/parser_js.py  JS/TS  -> IR (lightweight)
                                            ├─ _lib/explainer.py  IR -> plain-English steps
                                            ├─ _lib/graph.py      IR -> Mermaid flowchart
                                            └─ _lib/ai.py         IR + code -> AI summary + Big-O

The whole thing deploys to one Vercel project: the React app is served as static files and the Python analyzer runs as a serverless function at /api/explain. The backend uses only the Python standard library, so there are no dependencies to install and cold starts stay quick.

The AI part

_lib/ai.py asks a language model to summarize the code and estimate its complexity. It tries providers in order and always degrades gracefully:

1. Google Gemini — used if GEMINI_API_KEY is set (free tier, no card).
2. Pollinations — a free, keyless endpoint used when no Gemini key exists.
3. Built-in heuristic — if both are unavailable, a deterministic Big-O estimate (from loop nesting) is returned so the app never breaks.

Because of the fallback chain, CodeLensAI works out of the box for free. Adding a Gemini key just makes the AI summaries faster and more reliable.

---

Running locally

You need Node 18+ and Python 3.10+.

The simplest way to run the full stack (frontend + serverless API) locally is the Vercel CLI, which mirrors production exactly:

npm i -g vercel
vercel dev

Or run just the frontend against the deployed API:

cd frontend
npm install
npm run dev   # set VITE_API_BASE to your API origin if needed

Run the tests (no network required):

python tests/test_parser.py

---

Optional: add a free Gemini key

1. Create a key at https://aistudio.google.com/apikey (free, no credit card).
2. Add it to your Vercel project as an environment variable named GEMINI_API_KEY, then redeploy.

That's it — the app will automatically prefer Gemini for summaries.

---

Author

Built by Ishraq Basher — Computer Science @ NYU Tandon.